1. Top of page
  2. Abstract

Important progress has been made in the identification of specific environmental factors and estimation of hereditary components in bone density, quantitative ultrasound (QUS), and bone turnover indices. By contrast, the search for specific genes that regulate bone mass has progressed rather slowly, and the results are more difficult to interpret and reproduce. This article reviews the genetics of osteoporosis and problems plaguing genetic research. It is argued that the search for genes involved in the expression of osteoporotic phenotypes should be based on linkage studies in relatively homogeneous populations. Strategies for increasing the power of studies, such as making use of information from extended pedigrees and multivariate analysis, are discussed. With the advent of a comprehensive human genetic linkage map, a complete identification of genes for osteoporosis appears feasible. Understanding the genetic mechanisms and their interactions with environmental factors should allow more focused and cost-effective osteoporosis prevention and treatment strategies.


  1. Top of page
  2. Abstract

Genetics of osteoporosis

Osteoporosis is a consequence of accumulated disturbances in the skeletal homeostasis, resulting in reduced bone strength, deterioration in bone quality and ultimately low-trauma fracture. Although appreciation of the genetic contribution to osteoporosis susceptibility is a relatively recent development, it is well documented. From most twin studies in the 1980s and 1990s, the proportion of variance of bone mineral density (BMD; a measure of bone strength and a predictor of fracture) accounted for by genetic factors was around 65–92%.(1–9) This effect appeared to persist even in the late decades of life.(10) However, there are reservations about the reliability of this estimate, primarily because monozygotic (MZ; identical) twins are likely to share more similar environments than are dizygotic (DZ) twins, which could inflate the estimate of genetic effects.(11) Indeed, sibling studies have yielded lower estimates of heritability.(12–15) Nevertheless, although the exact strength of heritability could be debated, it is clear that variation in BMD between individuals is determined largely by genetic factors. Furthermore, there are common sets of genetic and environmental influences underlying the determination of BMD at various skeletal areas. For example, the genetic correlation between lumbar spine and femoral neck BMD is 0.64 compared with 0.57 for environmental factors.(9) This indicates that genetic factors that influence lumbar spine BMD are more likely to influence femoral neck BMD than are environmental factors. Estimates of heritability from the bivariate twin model suggest that approximately one-third of the genetic influence on variance of femoral neck BMD is mediated through the same gene or genes that influence the lumbar spine.(9)

BMD changes with age with rapid acquisition during adolescence, achievement of peak level possibly between the second and third decades of life, and progressive decline after the menopause (in women) and in elderly men. The evidence of genetic effects on change in BMD is, at present, less than convincing. Kelly et al. studying a group of 21 MZ and 19 DZ female twin pairs, aged 24–75 years, followed for up to 5 years (4 measurements per twin), found that the intraclass correlation in change in lumbar spine BMD was significantly higher in MZ pairs than in DZ pairs, with estimated heritability of 70%.(16) However, although the correlation was higher in MZ pairs than in DZ pairs in the rate of change in femoral neck BMD, this difference was not significant. Another study in 25 MZ and 21 DZ male twin pairs over 14 years at the distal radius showed no significant genetic effect on the rate of bone loss, although MZ twins had a higher correlation than DZ twins (0.61 vs. 0.41).(6) These studies would be consistent with stronger genetic effects on rate of change of BMD in sites with predominantly trabecular (lumbar spine) rather than cortical (femoral neck and wrist) bone.

Although the genetic contribution to bone loss has not been shown consistently, its role in the determination of variation in indices of bone formation and resorption is established. In twin studies in predominantly premenopausal women, genetic factors account for up to 65% of the intersubject variances of bone formation markers such as osteocalcin and C-terminal propeptide of type I collagen and bone-specific alkaline phosphatase.(17–19) A recent study in postmenopausal twins also found that genetic factors contribute to both bone formation and resorption markers.(20)

The relationship between bone turnover and BMD is well established. For example, each standard deviation higher level of bone specific alkaline phosphatase (BSAP) (6 μg/liter) was associated with a 4% lower level of BMD in both lumbar spine and femoral neck.(19) However, BMD and bone turnover may be regulated by different sets of genes. For example, bivariate genetic analysis indicated that genes influencing BSAP accounted for only 12% of the total genetic variance in lumbar spine BMD and 4% for femoral neck BMD.

Recent attention has focused on the role of bone structure and quality in the assessment of osteoporosis. Quantitative ultrasound (QUS) parameters, including broadband ultrasound attenuation (BUA) and speed of sound (SOS), have been proposed recently as measurements that reflect the quality aspects, for example, microarchitecture, of bone. These may provide additional information about osteoporotic fracture risk beyond that obtained from bone density.(21) Indeed, subjects with lower BUA at baseline have a higher risk of hip and vertebral fractures, possibly independent of BMD.(21–26) Estimates of heritability based on twin studies for age- and weight-adjusted BUA and SOS are 0.74 and 0.82, respectively.(27,28) Thus, these traits, presumably relating to bone quality, also are highly heritable. Under the presumption that QUS and BMD measure overlapping characteristics of bone and given the observation that both traits are determined genetically, then it could be expected that the traits also share common genetic factors. Indeed, bivariate genetic analysis indicated that the genetic correlation between BUA and BMD ranged between 0.43 and 0.51, whereas the environmental correlation ranged between 0.20 and 0.28.(28) These data are consistent with both common and specific sets of genes influencing QUS and BMD measurements.

Table Table 1.. Familial Relative Risk of Fracture for Various Age Groups and Familial Correlation
 Familial correlation in osteoporosis
  1. Based on expected prevalence of osteoporosis for age range 60–69, 70–79, and 80+ years of 16, 34, and 69%, respectively. Relative risk of fractures was derived from the Dubbo osteoporosis Epidemiology Study, among osteoporotic subjects at any fracture site, 3.5; hip fracture, 10.8; vertebral fracture, 8.7; and upper limb fractures, 2.8.(1)

60–69 years
Any fracture1.221.271.311.361.40
Hip fracture2.002.192.392.592.79
Vertebral fracture1.811.982.142.312.47
Upper limbs fracture1.
70–79 years
Any fracture1.221.271.311.361.40
Hip fracture1.631.761.892.022.16
Vertebral fracture1.541.651.761.871.98
Upper limbs fracture1.
80+ years
Any fracture1.
Hip fracture1.241.291.341.391.45
Vertebral fracture1.
Upper limbs fracture1.

Fracture is the ultimate consequence of osteoporosis. However, data on the heritability of fracture are scarce. Because BMD is the principal predictor of fracture risk and variation in BMD has a major hereditary component, it is possible to estimate the sibling relative risk for a given familial correlation of BMD. For example, for a familial correlation in BMD of 0.4, the relative risk of hip fracture in a sibling of a hip fracture subject can be as high as 1.9. This would be increased to 2.7 for a familial correlation of 0.8, particularly among those aged between 70 and 79 years (Table 1; the estimation of familial risk is derived from two parameters: relative risk of fracture among osteoporotic subjects and familial correlation in osteoporosis [as defined by femoral neck BMD]). These estimates are consistent with empirical data of a woman <80 years old with a maternal history of hip fracture, having a 2-fold (95% confidence interval, 1.4–2.9) increase in hip fracture compared with those without a maternal hip fracture.(29) They are also consistent with daughters of osteoporotic mothers or mothers with hip fracture having a significantly lower BMD at the lumbar spine and particularly at the femoral neck.(30–31) These studies collectively suggest the concept that variation in osteoporotic fracture risk has, in part, a genetic component.

In summary, data accumulated during the last three decades support the hypothesis that genetic factors are a major determinant of bone density and possibly variance of bone loss, bone formation, and bone resorption. Many common genetic factors appear to be responsible for the determination of BMD in various skeletal sites as well as bone architecture. Indeed, segregation analysis excluded the hypothesis that a single major gene controls bone density. Thus, as with many other complex multifactorial conditions, the most likely model for the genetics of osteoporosis is that the hereditary influence, which determines an individual's susceptibility to fracture, is mediated via several allelic loci, a few with large effects and the majority with moderate effects (Fig. 1). The recognition that genetic factors contribute to the development of osteoporosis has provided an incentive to identify and characterize specific loci or genes involved in determining bone-related phenotypes such as BMD, QUS, and bone turnover.

thumbnail image

Figure FIG. 1.. A hypothetical distribution of the number of osteoporosis genes by their effect size.

Download figure to PowerPoint


  1. Top of page
  2. Abstract

The identification of genes that contribute to the development of osteoporosis has potential scientific, therapeutic, and public health implications. Each such discovery could open new opportunities, through studying its action to improve our understanding of the interplay of genetic and environmental factors underlying osteoporosis. Most drugs work by binding to a specific protein, altering its activity to achieve a therapeutic outcome; thus, identification of a gene and, importantly, its protein product can lead to new targets for new therapeutic agents. Furthermore, in the future such knowledge of gene action may open the road to gene therapy.

In terms of public health, identifying genes can lead to new diagnostics for the early detection of osteoporosis risk and hence fracture risk. Once genes have been identified, it becomes possible to dissect the genetic variations that influence the intermediate risk factor traits and interact etiologically with environmental factors and the effects of other genes to explain the variability in susceptibility to osteoporosis.

Most studies on genetics of osteoporosis have addressed the concept of “main effect” rather than “interaction effect.” In genetic epidemiology, interaction is a phenomenon whereby effects of genetic factors vary according to levels of environmental exposure. Distinct gene alleles may produce different amounts or types of gene products that modify physiological and structural parameters, which in turn determine bone size, density, and structure as measured by BMD (or other modalities), which are themselves surrogates for fracture risk.

Currently, controversies exist as to whether preventive approaches in public health should be based on populations-at-large (low risk) or on selected (high risk) individuals. The former strategy advocates changes in diet, exercise, and lifestyle habits for the entire population, whereas the latter advocates selection of high-risk individuals for treatment. Although lifestyle modification has been proposed as an approach to prevent osteoporosis, there is insufficient evidence that such modifications even if effective have long-term efficacy in terms of fracture prevention or that such modifications could be implemented in populations at large with useful levels of compliance. On the other hand, the high-risk strategy is limited by the inability of present methods to detect such patients with high sensitivity and specificity leading to an unfavorable cost/benefit ratio of case finding of high-risk individuals in the general population. Given these conflicting viewpoints, the concept of gene-environmental interaction could be considered in designing a preventive program. Instead of identifying subjects with high risk (based on environmental factors) or focusing on all subjects in the population, it could be possible to identify subjects with both environmental exposure and high-risk genotype. This “genetic-environmental interaction preventative strategy” would be targeted at individuals whose risk of osteoporosis is elevated beyond that of overall high-risk genotype or environmental exposure. Identification of subjects with high-risk genotypes and environmental exposures for prevention and treatment could improve cost-effectiveness, as well as sensitivity and specificity of a preventative program.(32)

One area of study that has received little attention in the past is the issue of gene-gene interaction. A gene may have a modest effect on a phenotype but in combination with other genes may be responsible for a large variation in the phenotype. This area of investigation is complicated in the sense that large sample sizes and several genes are required in the analysis. However, with increasing identification of genes involving osteoporosis phenotypes, it is becoming possible to examine gene-gene interactions.


  1. Top of page
  2. Abstract

There are two basic strategies for searching genes that influence bone mass or any complex trait. In a candidate-gene approach, individual genes can be tested directly for their role in the determination of a trait of interest. A major disadvantage of this approach in osteoporosis is that multiple genes are likely to be involved in the regulation of the disease. Thus, when many genes are tested separately, some will be significant by chance alone. It is possible to correct for these associations by statistical adjustment, but by definition, candidate genes have a reasonable prior probability of being involved in disease susceptibility; therefore, correction for the number of genes tested should be conservative and over-correction could be counterproductive.

In a genome search approach, to ensure that all major loci involved in the control of a phenotype are identified, all genes are screened systematically using panels of microsatellite DNA markers uniformly distributed throughout the entire genome. This approach involves gathering a large number of related individuals thought to be segregating for genes that influence a trait and then tracing the putative parent-to-offspring cotransmission of variants (i.e., alleles or genotypes) at landmark spots (marker loci) along the genome. If alleles at a particular marker locus appear to segregate (or be transmitted) along with the presence of the trait or disease in question, it is possible to infer that a gene actually influencing the trait or disease is near or “linked” to the marker locus in question. The approximate location of a gene can be identified by indicating coinheritance of the bone phenotype with a particular marker. Any region identified by genome scanning typically spans 5–10 centi-Morgans (cM), and localization has to be progressively refined until a gene can be identified.


  1. Top of page
  2. Abstract

In each of the candidate gene and genome search approaches, genes can be identified based on a demonstration of a significant association and/or linkage relationship. In the context of bone mass, association analysis tests whether BMD and an allele show correlated occurrence in a population, whereas linkage analysis tests whether the genes show correlated transmission within pedigrees.

Linkage analysis involves the proposition of a model to account for the inheritance pattern of phenotype observed in a pedigree. Specifically, it determines whether the phenotypic locus is transmitted with genetic markers of known chromosomal position. Linkage with a phenotypic locus can be tested marker by marker (two-point analysis) or by a set of linked markers (multipoint analysis). The evidence of linkage is evaluated according to the LOD score.(33) The LOD score is a likelihood ratio testing the hypothesis of linkage against the null hypothesis of no linkage for different genetic distances (e.g., recombination fractions) between the phenotype locus and the marker locus. Based on a certain threshold of the LOD score, linkage can be accepted or excluded. It has been proposed to interpret a LOD score of 3 or higher as evidence of linkage, whereas a LOD score of 2–3 is an indication of “suggestive linkage.”(34) A LOD score of 3 implies that approximately 95% of declared linkages would be true.

There are two general approaches to linkage analysis: parametric pedigree analysis and nonparametric methods. Parametric pedigree analysis involves tracing cosegregation and recombination phenomena between observed marker alleles and unobserved putative-trait influencing alleles among members of large pedigrees. This method requires a specification of gene frequency and penetrance of the marker. In contrast, nonparametric linkage analysis examines the genetic factors influencing a phenotype without any phenotype-genotype model specification. The latter model is particularly appropriate to osteoporosis, because there are no known major gene effects that can be inferred from segregation analysis. One of the most popular nonparametric models is allele sharing, which assesses the number of marker alleles shared at a particular locus among pairs of relatives manifesting the same trait. Pedigree and allele-sharing analysis approaches focus on the detection of individual loci (much like many candidate gene analyses). Hence, they are less useful for analysis of multigenic traits.

Association (also a nonparametric model) is a phenomenon whereby a certain gene allele or locus is found at higher frequency in those with a particular phenotype. For a quantitative trait such as BMD, an “association” is suggested by a statistically significant difference between alleles with respect to the mean of BMD. Association studies are thus simpler than linkage studies. However, their limitations and potential confounders must be recognized.

It is important to stress that it is possible to find linkage without association. This would be the case when there are many trait-causing loci in a population so that association with any particular allele is weak. On the other hand, a “true” association can be shown in the absence of a significant linkage. This can be seen when an allele explains a minor proportion of the total variance for a trait so that, although the allele may occur more often in affected individuals, it has poor discriminant power for phenotype status within a pedigree. Thus, a significant association with a marker gene does not confirm that the marker gene and the trait locus are the same. Furthermore, true association caused by linkage disequilibrium can yield seemingly contradictory results in different populations, because linkage disequilibrium depends on each population's genetic history and possibly other genetic traits in different populations. Based on these considerations, a trait might show one direction of association with one allele in a population, in the opposite effect in a second population, or no effect in a genetically mixed population.

It should be mentioned here that in order to minimize the effect of population admixture, family based association methods, such as the transmission linkage disequilibrium test (TDT), have been developed. The sampling unit in this test consists of two parents with an affected offspring; parental alleles not transmitted to affected offspring are used as controls. Under certain circumstances, the TDT test can be more powerful than the sibpair linkage test.(35)


  1. Top of page
  2. Abstract

Using the candidate gene approach, Morrison et al. first reported an association of noncoding region polymorphisms in the vitamin D receptor (VDR) gene with a serum bone turnover marker, osteocalcin.(36) Subsequently, polymorphisms of this gene were found to be in linkage with bone density in a twin study.(37) It was shown further in an association analysis that higher lumbar spine BMD was associated with the Bsm-1 b allele, and the association was codominant such that subjects with bb genotype had higher lumbar spine BMD than Bb subjects, who had higher BMD than subjects with BB genotype. Genotype errors later identified suggest a weaker effect.(38) Although one twin study showed an association but not linkage of the VDR polymorphisms with bone density, another larger study found neither association nor linkage between the VDR polymorphisms and bone mass in their twin population.(39,40) A range of carefully performed population studies has found an association or no association, some even in the opposite direction (of effect) to the original reports.(41) Although some of the differences could be attributed to small sample sizes, the overall result indicates a significant effect but weaker than originally suggested.(42,43)

By using the same candidate-gene approach, other genes with similar size effects for allelic differences have been identified: genes for collagen Iα1 and receptors for estrogen, calcitonin, interleukin-6, and interleukin-1 as well as genes for HS-glycoprotein, transforming growth factor β1 (TGF-β1), and apoliprotein E.(44–55) However, like the situation with the VDR studies, subsequent studies also have yielded contradictory results.(56) It is important to recognize that an association could be found because the genetic marker is not within the etiological gene, but rather may only be in linkage disequilibrium with an allele from another nearby genetic locus with the trait-causing allele. Importantly, any observed association may be caused by a random, artifact of population admixture, with dominating founder effects.(57)

Any genetic relationship observed could be modified by interaction of these genes with the genetic or environmental background. For instance, dietary calcium intake, a possible contributor to BMD, varies widely (up to 1000 mg/d) between Asian and white populations and between studies. This mechanism was examined in the studies of Ferrari et al. and Krall et al.(58,59) In the former study, subjects who were carriers of Bb genotype of the VDR responded to calcium intake, that is, no significant bone loss, whereas subjects with bb or BB experienced significant bone loss regardless of dietary calcium intake. In the second study, at low dietary calcium intake, subjects with BB genotype responded to calcium supplementation, whereas the bb or Bb subjects did not.(59) Importantly, many of these studies have been based on relatively small sample sizes and thus may not have sufficient power to detect a true association or to exclude one with reasonable confidence. Moreover, it is worth noting that almost all studies so far have been based on association, not linkage, analysis.

Two complex and opposing processes of formation and resorption regulate the amount of bone present in any stage of life. Both processes are potentially influenced by different genetic factors. This dynamic system creates enormous potential for genetic variations to impact on bone phenotype. Yet, identification of genes for osteoporosis is plagued by this very factor: that the effect of any particular gene may be obscured or confounded by the effects of others yet to be identified. Analysis of each candidate in isolation of the others may amount to testing every gene on the human genome, an endeavor fraught with statistical problems relating to false-positive results. In addition, there is likely to be a great deal of heterogeneity both with respect to the genes that predispose one to or protect against osteoporosis and with respect to the environment that may induce or reduce susceptibility to osteoporosis. Hence, finding appropriately homogenous case and control groups will be problematic.


  1. Top of page
  2. Abstract

Populations, sampling units, and phenotypes

Given the problems of the candidate gene approach and association studies, the search for osteoporosis genes would be more efficiently based on genomic screening and linkage analysis in related individuals. In fact, linkage analysis and genomic screening have become somewhat standard tools in the search for genes involved in other presumed polygenic disorders such as asthma, obesity, diabetes, and pulmonary hypertension.(61–64) These studies have variously provided good experience that should be considered in any future genetic studies of osteoporosis. Among others, issues of study population, sampling units, and analysis deserve special attention to avoid previously encountered pitfalls.

Genetic linkage analyses and genomewide searches require a large number of families with multiple genetically related individuals, in whom the trait of interest segregates. However, increasing the numbers of families will introduce problems associated with heterogeneity of environmental exposures and genetic or ethnic backgrounds. Hence, the effect of one gene may be “neutralized” by the effect of others (i.e., its effect is not constant and detectable or even present in all the individuals in the sample.) The study population therefore should be as homogenous as possible to avoid effects of admixture. This may involve studies of relatively isolated populations or populations that do not undergo much migration, because these are more likely to have more restricted gene pools and thus are genetically more homogenous. However, in such restricted samples, genes contributing in other populations are likely to have different allelic frequencies so that their role would be overlooked. It may be more efficient to study younger populations (e.g., at the age of peak bone mass or younger), because this is the period in which the differences in bone phenotype are likely to be genetic in origin with less effects from aging and environmental exposure.

The most common sampling strategy for genomic screening is to study samples of well-defined families or sibpairs. This strategy is based on the notion that if a specific gene contributes to the disease in both members of a sibpair, then their genotypes at that disease gene and genetic markers nearby should be more similar than expected on the basis of chance, or in the case of quantitative phenotype such as bone density, sibpairs that share the same marker genotypes will have more similar phenotype.(65) The similarity of genotypes is based on the concept of identical by descent (IBD). Two genes are IBD if one is a physical copy of the other or they both are physical copies of the same ancestral gene. Two siblings can share zero, one, or two parental IBD alleles of any locus, and under random segregation, the probability of this sharing is 25, 50, and 25%, respectively (Fig. 2). Thus, the test for linkage of a marker to a disease locus compares the observed IBD distribution to that expected assuming no linkage. For a quantitative trait, this involves the regression of squared intrapair differences against the proportion of genes shared IBD.(66) This analysis requires no specification of mode of inheritance and is thus referred to as a “nonparametric” method.

Most past studies on candidate genes have been focused on BMD. Although BMD is an important phenotype in osteoporosis, other phenotypes such as QUS measurements and bone size also are relevant and need to receive more genetic consideration. Studies on intermediate phenotypes such as bone formation and bone resorption markers also may be informative. Genetic analyses of these low-level phenotypes may be simpler if their greater proximity to the genetic substrate reduces the number of genes and exogenous factors involved. Such studies of intermediate phenotypes also may provide better understanding of the underlying physiological and biochemical determinants of the bone phenotype.

Statistical analyses

The nonparametric method, unfortunately, has low power. To have an 80% chance for detecting a gene accounting for 10% of the variance of BMD, a sample size of more than 80,000 sibpairs is required. Even if the gene has a large effect (heritability of 40%), the sample size is still in the range of 3600–5100 pairs.(67) These numbers and associated costs are almost beyond the reach of most investigations. In an effort to reduce the number of siblings required, a number of investigators have proposed alternative sampling plans. One of such plans is to select sibpairs who are extremely discordant for a trait.(68–71) The reason this approach is powerful is that for a complex polygenic trait such as bone density, extremely discordant pairs are unlikely to share any polygenes contributing to the phenotype. For example, to have an 80% chance of detecting a locus with a heritability of 10% and the predisposing allele frequency of 0.10 with no other genetic or common environmental influences, a sample of 1647 discordant sibpairs (one sibling is in the 1st decile and the other is in the 30th decile) would be required. However, the number of sibpairs screened for such a criterion would be very large (19,120), because the probability of obtaining a pair whose siblings are in the 1st and 30th deciles is low (Table 2).

thumbnail image

Figure FIG. 2.. An illustration of the concept of IBD. Individuals 3 and 4 share 2 alleles IBD; individuals 4 and 5 share 1 allele IBD; and individuals 5 and 6 share no alleles IBD. In a randomly mating population, the proportions of sibpairs who share zero, one, and two alleles IBD at any locus are 25, 50, and 25%, respectively; and the expected mean proportion of alleles that they share IBD is 50%.

Download figure to PowerPoint

Table Table 2.. Number of Extremely Discordant Sibpairs Required to be Analyzed and Screened to Detect Linkage, Under Additive, Dominant, and Recessive Models
 Heritability of the locus
Allele frequency0.10.20.3
  1. Pairs that are extremely discordant in this model include pairs with one in the top 10% and the other in the opposite extreme 30%. The figures were adapted from Risch and Zhang.71 Numbers in brackets are number of sibpairs required to be screened. Assumptions: zero residual correlation, power of 80% and significance level of 0.0001 (corresponding to a LOD score of 3).

Additive model
0.12544 (9575)507 (2180)185 (910)
0.32967 (11,075)596 (2526)217 (1056)
0.53377 (12,518)707 (2941)263 (1239)
Dominant model
0.12476 (9294)471 (2023)168 (828)
0.33352 (12,307)698 (2843)263 (1192)
0.54879 (17,513)1289 (4938)596 (2421)
Recessive model
0.119,991 (68,430)19,001 (65,085)18,996 (65,068)
0.32591 (9434)622 (2462)282 (1200)
0.52679 (9855)524 (2154)193 (889)

The sibpair design is simple, in which a set of only two related individuals is required. However, it is intuitive that larger sibships can provide more information than two individual sibpairs. For a sibship with m siblings, it is possible to form m(m − 1)/2 sibpairs, although only m − 1 are independent.(72) A study with a set of sibships of size 2 requires measurement of phenotype for 2 individuals to obtain one sibling pair, thus the yield of sibling pairs per phenotypic measurement is ½. But for a sibship with m siblings the ratio of sibling pairs to phenotypic measurements taken is (m − 1)/2. Thus, the relative efficiency of large sibships versus randomly selected pairs is m − 1. Todorov et al. (1997) have shown that large sibships can be a cost-effective alternative to the use of sampling of sibling pairs.(73) However, sibpairs formed from the formulation of m(m − 1)/2 are not independent and the significance level (type I error) can be inflated. However, with the recent advance in statistical genetics, particularly the variance component approach, this problem can be resolved by maximizing the likelihood of a sibship jointly on all members in the sibship.(74) In fact, with the variance component method, it is possible to increase the power of a study by including all individuals in any extended pedigree in the analysis.(75)

Statistical significance level and power

In a typical epidemiological or clinical study, there are two types of errors that can be associated with significance testing: a true null hypothesis (e.g., no difference between groups) can be rejected incorrectly and a false null hypothesis can fail to be rejected. The former error is called a type I error (usually denoted by α) and the latter error is called a type II error (denoted by β). The probability of finding a false linkage at one locus is called a locus-specific or nominal significance level and is usually set at 0.05 or 5%. However, this criterion is not stringent enough, because in a genomic scan study, many loci are tested for linkage, which increases the type I error such that many false-positives (e.g., false linkages) can be expected by chance. It has been proposed that the locus-specific significance level be determined approximately from the equation α = αg/(23 = 123Zα2), where αg is the probability of finding a false linkage at any of the many loci tested (genomewide significance level)and Zα is the standard normal deviate associated with the significance level α.(34) For example, if αg = 0.05, then a locus-specific significance level of α = 0.000018 is required.

A type II error is only an error in the sense that an opportunity to reject correctly the null hypothesis was lost. It is not an error in the sense that an incorrect conclusion was drawn because no conclusion is drawn when the null hypothesis is not rejected. Power, defined as 1 – β is one of the key issues in any linkage study. If the power of a study is low, there is a good chance that the study's results will be inconclusive. However, in a genomic screening study, there are actually three types of power: locus-specific power is the probability of detecting an individual locus, genomewide power is the probability of detecting any of the k loci considered, and studywise power is the probability of detecting all k loci associated with a trait. Let the locus-specific power be 1 – βi(for an ith locus, i = 1, 2, 3, …, k), and then the genomewide power is 1 – β1 × β2 × β3 × … × βk, and the study power is (1 – β1) × (1 – β2) × 1 – β3) × … × (1 – βk). Intuitively, the genomewide power is more appropriate in a genomic scan study. Thus, for a traditional 1 – βi of 0.8 and with an assumed 5 loci involved, the genomewide power is 1 – (0.2)5 = 0.9997; that is, the chance of detecting linkage with any locus is considerably higher than the chance of detecting a specific locus. However, this calculation does not imply that small sample size in a genomic study is the design of choice, rather it suggests that studies with relatively small sample size still may be useful because the chance of detecting a linkage can be quite high.

Among strategies for increasing power, such as the selection of sampling unit, other strategies relating to data analysis also are worth considering. Osteoporosis is characterized by many phenotypes relating to bone density and bone quality. Past association studies have been based primarily on univariate analysis, in which each phenotype was analyzed separately. As in the case of multiple loci, multiple univariate statistical analyses ultimately increase the type I error rate, under the complete null hypothesis of no linkage. Procedures for correction such as Bonferroni adjustment can be applied; however, the results may be too conservative, because the phenotypes are likely correlated. Indeed, if a study examines m traits, with the average correlation of r, then the number of equivalent independent traits is (1 – r)(m − 1) + 1, which is almost always less than m the adjustment factor.(76) Instead of adjusting the significance level of univariate analyses, a multivariate analysis may be preferable. It has been estimated that linear combination of many traits into factor scores is more powerful than the use of multivariate or mean phenotypic data.(77–79)

Other strategies to increase the chance of finding genes for osteoporosis may include a study of intermediate phenotypes such as bone formation and bone resorption markers and hormones also may be informative, because genetic analyses of these low-level phenotypes may be simpler if their greater proximity to the genetic substrate reduces the number of genes and exogenous factors involved. Such studies of intermediate phenotypes also may provide a better understanding of the underlying physiological and biochemical determinants of the bone phenotype. Typing a greater proportion of parents, increasing marker heterozygosity, increasing marker density, and increasing measurement reliability are also worth consideration.(80,81)

Animal models

Important developments in the understanding of the genetics of obesity and deafness in the human population have come from studies of animal models.(82,83) Recent developments of comprehensive genetic maps along with rapid progress toward a physical map of the mouse genome have provided tools to aid in the identification of genes underlying interesting new mutants.(84) They also enable investigators to identify genes mapped in the mouse genome that may be candidates for human disease–causing genes and vice versa.

The mouse mutant catalogue documents around 1000 mutations, which is only 1–2% of the total number of mammalian genetic loci.(85) Therefore, there is scope for increasing both the breadth (new loci) and depth (new mutations of known loci) of the current mouse mutant resource. Two approaches can be considered toward the generation and recovery of new mutations: the genotype-driven and phenotype-driven mutagenesis. A variety of routes are available to introduce new mutations into the mouse germline. These can involve classic transgenic approaches whereby constructs introduced into the genome by pronuclear injection lead to insertional mutagenesis. Alternatively, homologous recombination in embryonic stem cells can be used to introduce new mutations into known genes.(86) This approach has the advantage of easy identification of an underlying gene that has been disrupted using the inserted selectable marker, but it must be married to an intensive phenotype per assessment in order to identify “informative” mutations, which may not be intuitively obvious from the genetic locus involved.

The phenotype-driven approach focuses on mutagenesis procedures that emphasize the recovery of new phenotypes without assumptions about the nature of the underlying genes or biological pathways. For example, N-ethyl-N-nitrosourea has the potential to increase enormously the range of the mutant resource, it contrasts with the genotype-driven approach in that identification of the gene underlying any mutation is not trivial. These two approaches are in fact complementary rather than exclusive alternatives. These developments and approaches in mouse genomics will allow efficient identification of genes involved with human diseases. One major advantage of using an animal model is that it is possible to control for the heterogeneity of environmental factors in animals, which is otherwise impossible in human studies. For example, Klein et al. have used quantitative trait loci (QTL) analysis in genetically distinct mice that were raised in strictly controlled environments and have preliminarily identified 10 chromosomal regions linked to peak bone mass.(87) Such an animal model will be an increasingly valuable tool of choice for the identification and analysis of new genes and biological pathways of osteoporosis coming from the genetic research of osteoporosis in the genomic and postgenomics era.

In summary, from the clinical as well as economic points of view, aggressive strategies to detect osteoporosis genes appear to be warranted. Operational research is needed to define the roles of genes, environments, and their interactions more clearly to develop more effective prevention programs for osteoporosis. Comprehensive human genetic linkage maps in conjunction with linkage studies with a large number of markers covering most of the chromosomal length of the human genome as well as advances in analytical methods in combination with animal models will facilitate identification of osteoporosis genes. Understanding their mechanisms and interactions with environmental factors should allow more focused and cost-effective osteoporosis prevention and treatment strategies.


  1. Top of page
  2. Abstract
  • 1
    Nguyen TV, Sambrook P, Kelly P, Jones G, Lord S, Freund J, Eisman J 1993 Prediction of osteoporotic fractures by postural instability and bone density. BMJ 307: 11111115.
  • 2
    Hui SL, Slemenda CW, Johnston CJ 1988 Age and bone mass as predictors of fracture in a prospective study. J Clin Invest 81: 18041809.
  • 3
    Melton LJ III, Atkinson EJ, O'Fallon WM, Wahner HW, Riggs BL 1993 Long term fracture prediction by bone mineral assessed at different skeletal sites. J Bone Miner Res 8: 12271233.
  • 4
    Cummings SR, Black DM, Nevitt MC, Browner W, Cauley J, Ensrud K, Genant HK, Palermo L, Scott J, Vogt TM 1993 Bone density at various sites for prediction of hip fractures. Lancet 341: 7275.
  • 5
    Dequeker J, Nijs J, Verstraeten A, Geusens P, Gevers G 1987 Genetic determinants of bone mineral content at the spine and radius. Bone 8: 207209.
  • 6
    Christian JC, Yu P, Slemenda CW, Johnston CC 1989 Heritability of bone mass: A longitudinal study in aging male twins. Am J Hum Genet 44: 429433.
  • 7
    Young D, Hopper JL, Nowsen CA, Green RM, Sherwin J, Kaymacki B, Smid M, Guest GS, Larkins RG, Wark JD 1995 Determinants of bone mass in 10 to 26 years old females: A twin study. J Bone Miner Res 10: 558567.
  • 8
    Pocock NA, Eisman JA, Hopper JL, Yeates MG, Sambrook PN, Eberl S 1987 Genetic determinants of bone mass in adults: a twin study. J Clin Invest 80: 706710.
  • 9
    Nguyen TV, Howard GM, Kelly PJ, Eisman JA 1998 Bone mass, lean mass and fat mass: same genes or same environmemts. Am J Epidemiol 147: 316.
  • 10
    Flicker L, Hopper JL, Rogers L, Kaymacki B, Green RM, Wark JD 1995 Bone mineral density determinants in elderly women: a twin study. J Bone Miner Res 10: 16071613.
  • 11
    Slemenda CW, Christian JC, Williams CJ, Norton JA, Johnston CC 1991 Genetic determinants of bone mass in adult women: A reevaluation of the twin model and the potential importance of gene interaction on heritability estimates. J Bone Miner Res 6: 561567.
  • 12
    Lutz J, Tesar R 1990 Mother-daughter pairs: spinal and femoral bone densities and dietary intakes. Am J Clin Nutr 52: 872877.
  • 13
    Evans RA, Marel GM, Lancaster EK, Kos S, Evans M, Wong SYP 1988 Bone mass is low in relatives of osteoporotic patients. Ann Intern Med 109: 870873.
  • 14
    Tylavsky FA, Bortz AD, Hancock RL, Anderson JJ 1989 Familial resemblance of radial bone mass between pre-menopausal mothers and their college age daughters. Cacilf Tissue Int 45: 265272.
  • 15
    Krall EA, Dawson-Hughes B 1993 Heritability and life-style determinants of bone mineral density. J Bone Miner Res 8: 19.
  • 16
    Kelly PJ, Nguyen TV, Hopper JL, Pocock N, Sambrook PN, Eisman JA 1993 Change in axial bone density with age: A twin study. J Bone Miner Res 8: 1119.
  • 17
    Kelly PJ, Hopper JL, Macaskill GT, Pocock NA, Sambrook PN, Eisman JA 1991 Genetic factors in bone turnover. J Clin Endocrinol Metab 72: 808813.
  • 18
    Tokita A, Kelly, PJ, Nguyen TV, Sambrook PN, Eisman JA 1994 Genetic influences on type I collagen synthesis and degradation: Further evidence for genetic regulation of bone turnover. J Clin Endocrinol Metab 78: 14611466.
  • 19
    Harris M, Nguyen TV, Kelly PJ, Howard GM, Eisman JA 1998 Genetic and environmental correlations between bone formation and bone mineral density: A twin study. Bone 22: 141145.
  • 20
    Garnero P, Arden NK, Griffiths G, Delmas PD, Spector TD 1996 Genetic influence on bone turnover in postmenopausal twins. J Clin Endocrinol Metab 81: 140146.
  • 21
    Kaufman JJ, Einhorn TA 1993 Perspectives: Ultrasound assessment of bone. J Bone Miner Res 8: 517525.
  • 22
    Bauer DC, Gluer CC, Cauley JA, Vogt TM, Ensrud KE, Genant HK, Black DM. 1997 Bone ultrasound predicts fractures strongly and independently of densitometry in older women: A prospective study. Arch Intern Med 157: 629634.
  • 23
    Gluer CC, Cummings SR, Bauer DC, Stone K, Pressman A, Mathur A, Genant HK 1996 Osteoporosis: Association of recent fractures with quantitative US findings. Radiology 199: 725732.
  • 24
    Bauer DC, Gluer CC, Genant HK, Stone K. 1995 Quantitative ultrasound and vertebral fracture in postmenopausal women. J Bone Miner Res 10: 353358.
  • 25
    Ross P, Huang C, Davis J, Imose K, Yates J, Vogel J, Wasnich R 1995 Predicting verterbal deformity using bone densitometry at various skeletal sites and calcaneus ultrasound. Bone 16: 325332.
  • 26
    Hans D, Dargent-Molina P, Schott AM, Sebert JL, Cormier C, Kotzki PO, Delmas PD, Pouilles JM, Breart G, Meunier PJ 1996 Ultrasonographic heel measurements to predict hip fracture in elderly women: The EPIDOS prospective study. Lancet 348: 511514.
  • 27
    Arden NK, Baker J, Hogg C, Baan K, Spector TD 1996 The heritability of bone mineral density, ultrasound of the calcaneus and hip axis length: A study of postmenopausal twins. J Bone Miner Res 11: 530534.
  • 28
    Howard GM, Nguyen TV, Harris M, Kelly PJ, Eisman JA 1998 Genetic and environmental contributions to the association between quantitative ultrasound and bone mineral density measurements: A twin study. J Bone Miner Res 13: 13181327.
  • 29
    Cummings SR, Nevitt MC, Browner WS, Stone K, Fox K, Ensrud K, Cauley J, Black D, Vogt TM 1995 Risk factors for hip fractures in white women. New Engl J Med 332: 767773.
  • 30
    Seeman E, Hopper JL, Bach LA, Cooper ME, Parkinson E, MacKay J, Jerums G 1989 Reduced bone mass in daughters of women with osteoporosis. N Engl J Med 320: 554558.
  • 31
    Seeman E, Tsalamandris C, Formica C, Hopper JL, McKay J 1994 Reduced femoral neck bone density in the daughters of women with hip fractures: The roles of low peak bone density in the pathogenesis of osteoporosis. J Bone Miner Res 9: 739743.
  • 32
    Khoury MJ 1997 Genetic epidemiology and the future of disease prevention and public health. Epidemiol Rev 19: 175180.
  • 33
    Morton NE 1955 Sequential tests for the detection of linkage. Am J Hum Genet 7: 731749.
  • 34
    Lander E, Kruglyak L 1995 Genetic dissection of complex traits: guidelines for interpreting and reporting linkage results. Nat Genet 11: 241247.
  • 35
    Risch N, Merikangas K 1996 The future of genetic studies of complex human diseases. Science 273: 15161517.
  • 36
    Morrison NA, Yeoman R, Kelly PJ, Eisman JA 1992 Contribution of trans-acting factor alleles to normal physiological variability: Vitamin D receptor gene polymorphisms and circulating osteocalcin. Proc Natl Acad Sci U S A 89: 66656669.
  • 37
    Morrison NA, Qi JC, Tokita A, Kelly P, Croft L, Nguyen TV, Sambrook PN, Eisman JA 1994 Prediction of bone density by vitamin D receptor alleles. Nature 367: 284287.
  • 38
    Morrison NA, Qi JC, Tokita A, Kelly P, Croft L, Nguyen TV, Sambrook PN, Eisman JA 1997 Prediction of bone density by vitamin D receptor alleles: Corrections. Nature 387: 106.
  • 39
    Spector TD, Keen RW, Arden NK, Morrison NA, Major PJ, Nguyen TV, Kelly PJ, Baker JR, Sambrook PN, Lanchbury JS 1995 Influence of vitamin D receptor genotype on bone mineral density in postmenopausal women: A twin study in Britain. BMJ 310: 13571360.
  • 40
    Hustmyer FG, Peacock M, Hui S, Johnston CC, Christian J 1994 Bone mineral density in relation to polymorphism at the vitamin D receptor gene locus. J Clin Invest 94: 21302134.
  • 41
    Uitterlinden AG, Pols HAP, Burger H, van Daele PLA, van Duijn CM, Hofman A, Birkenhager JC, van Leeuwen JPTM 1996 Haplotypes at the vitamin D receptor gene locus are associated with bone mineral density. J Bone Miner Res 11: 12411248.
  • 42
    Nguyen TV, Kelly PJ, Morrison NA, Sambrook PN, Eisman JA 1994 Vitamin D receptor genotypes in osteoporosis. Lancet 344: 15801581.
  • 43
    Cooper GS, Umbach DM 1996 Are vitamin D receptor polymorphisms associated with bone mineral density? A meta analysis. J Bone Miner Res 11: 18411849.
  • 44
    Grant SFA, Reid DM, Blake G, Herd R, Fogelman I, Ralston SH 1996 Reduced bone density and osteoporotic fracture associated with a polymorphic Sp1 binding site in the collagen type I a 1 gene. Nature Genetics 14: 203205.
  • 45
    Qi JC, Morrison NA, Nguyen TV, White CP, Kelly PJ, Sambrook PN, Eisman JA 1994 Estrogen receptor genotypes and bone mineral density in women and men. J Bone Miner Res 10: S170.
  • 46
    Sano M, Inoue S, Hosoi T, Ouchi Y, Emi M, Shiraki M, Orimo H 1995 Association of estrogen receptor dinucleotide repeat polymorphism with osteoporosis. Biochem Biophys Res Commun 217: 378383.
  • 47
    Kobayashi S, Inoue S, Hosoi T, Ouchi Y, Shiraki M, Orimo H 1996 Association of bone mineral density with polymorphism of the estrogen receptor gene. J Bone Miner Res 11: 306311.
  • 48
    Masi L, Becherini L, Colli E, Gennari L, Mansani R, Falchetti A, Becorpi AM, Cepollaro C, Gonnelli S, Tanini, A, Brandi ML 1998 Polymorphisms of the calcitonin receptor gene are associated with bone mineral density in postmenopausal Italian women. Biochem Biophys Res Commun 248: 190195.
  • 49
    Taboulet J, Frenkian M, Frendo JL, Feingold N, Jullienne A, de Vernejoul MC 1998 Calcitonin receptor polymorphism is associated with a decreased fracture risk in post-menopausal women. Hum Mol Genet 7: 21292133.
  • 50
    Houston LA, Grant SF, Reid DM, Ralston SH 1996 Vitamin D receptor polymorphism, bone mineral density, and osteoporotic vertebral fracture: Studies in a UK population. Bone 18: 249252.
  • 51
    Keen RW, Woodford-Richens KL, Lanchbury JS, Spector TD 1998 Allelic variation at the interleukin-1 receptor antagonist gene is associated with early postmenopausal bone loss at the spine. Bone 23: 367371.
  • 52
    Eichner JE, Friedrich CA, Cauley JA, Kamboh MI, Gutai JP, Kuller LH, Ferrell RE 1990 Alpha 2-HS glycoprotein phenotypes and quantitative hormone and bone measures in postmenopausal women. Calcif Tissue Int 47: 345349.
  • 53
    Zmuda JM, Eichner JE, Ferrell RE, Bauer DC, Kuller LH, Cauley JA 1998 Genetic variation in alpha 2HS-glycoprotein is related to calcaneal broadband ultrasound attenuation in older women. Calcif Tissue Int 63: 58.
  • 54
    Langdahl BL, Knudsen JY, Jensen HK, Gregersen N, Eriksen EF 1997 A sequence variation: 713-8delC in the transforming growth factor-beta 1 gene has higher prevalence in osteoporotic women than in normal women and is associated with very low bone mass in osteoporotic women and increased bone turnover in both osteoporotic and normal women. Bone 20: 289294.
  • 55
    Cauley JA, Zmuda JM, Yaffe K, Kuller LH, Ferrell RE, Wisniewski SR, Cummings SR 1999 Apolipoprotein E polymorphism: A new genetic marker of hip fracture risk—the study of osteoporotic fractures. J Bone Miner Res 14: 11751181.
  • 56
    Zmuda JM, Cauley JA, Ferrell RE 1999 Recent progress in understanding the genetic susceptibility to osteoporosis. Genet Epidemiol 16: 356367.
  • 57
    Rosenberg L, Kidd KK 1977 HLA and disease susceptibility: A primer. N Engl J Med 297: 10601062.
  • 58
    Ferrari S, Rizzoli R, Manen D, Slosman D, Bonjour JP 1998 Vitamin D receptor gene start codon polymorphisms (FokI) and bone mineral density: Interaction with age, dietary calcium, and 3[prime]-end region polymorphisms. J Bone Miner Res 13: 925930.
  • 59
    Krall EA, Parry P, Lichter JB, Dawson-Hughes B 1995 Vitamin D receptor alleles and rates of bone loss: Influences of years since menopause and calcium intake. J Bone Miner Res 10: 978984.
  • 60
    Nguyen TV, Sambrook PN, Eisman JA 1998 Bone loss, physical activity, and weight chang in elderly women: The Dubbo osteoporosis epidemiology study. J Bone Miner Res 13: 14581467.
  • 61
    Drazen JM, Silverman EK 1997 Genetics of asthma. Am J Respir Crit Care Med 156: S69S71.
  • 62
    Norman RA, Tataranni PA, Pratley R, Thompson DB, Hanson RL, Prochazka M, Baier L, Ehm MG, Sakul H, Foroud T, Garvey WT, Burns D, Knowler WC, Bennett PH, Bogardus C, Ravussin E 1998 Autosomal genomic scan for loci linked to obesity and energy metabolism in Pima Indians. Am J Hum Genet 62: 659668.
  • 63
    Elbein SC, Hoffman MD, Teng K, Leppert MF, Hasstedt SJ 1999 A genome-wide search for type 2 diabetes susceptibility genes in Utah Caucasians. Diabetes 48: 11751182.
  • 64
    Nichols WC, Koller DL, Slovis B, Foroud T, Terry VH, Arnold ND, Siemieniak DR, Wheeler L, Phillips JA 3rd, Newman JH, Conneally PM, Ginsburg D, Loyd JE 1997 Localization of the gene for familial primary pulmonary hypertension to chromosome 2q31-32. Nat Genet 15: 277280.
  • 65
    Penrose LS 1935 The detection of autosomal linkage in data which consist of pairs of brothers and sisters of unspecified parentage. Ann Eugenics (London) 6: 133138.
  • 66
    Haseman JK, Elston RC 1972 The investigation of linkage between a quantitative trait and a marker locus. Behav Genet 2: 319.
  • 67
    Blackwelder WC, Elston RC 1995 A comparison of sib-pair linkage tests for disease susceptibility loci. Genet Epidemiol 2: 8597.
  • 68
    Cardon LR, Fulker DW 1994 The power of interval mapping of quantitative trait loci, using selected sib pairs. Am J Hum Genet 55: 825833.
  • 69
    Carey G, Williamson J 1991 Linkage analysis of quantitative traits: increased power by using selected samples. Am J Hum Genet 49: 786796.
  • 70
    Eaves L, Meyer J 1994 Locating human quantitative trait loci: guidelines for the selection of sibling pairs for genotyping. Behav Genet 24: 443455.
  • 71
    Risch NJ, Zhang H 1996 Mapping quantitative trait loci with extreme discordant sib pairs: Sampling considerations. Am J Hum Genet 58: 836843.
  • 72
    Hodge SE 1984 The information contained in multiple sibling pairs. Genet Epidemiol 1: 109122.
  • 73
    Todorov AA, Province MA, Borecki IB, Rao DC 1997 Trade-off between sibship size and sampling scheme for detecting quantitative trait loci. Hum Hered 47: 15.
  • 74
    Almasy L, Blangero J 1998 Multipoint quantitative-trait linkage analysis in general pedigrees. Am J Hum Genet 62: 11981211.
  • 75
    Williams JT, Blangero J 1999 Comparison of variance components and sibpair-based approaches to quantitative trait linkage analysis in unselected samples. Genet Epidemiol 16: 113134.
  • 76
    Morton NE 1998 Significance levels in complex inheritance. Am J Hum Genet 62: 690697.
  • 77
    Allison DB, Thiel B, St Jean P, Elston RC, Infante MC, Schork NJ 1998 Multiple phenotype modeling in gene-mapping studies of quantitative traits: Power advantages. Am J Hum Genet 63: 11901201.
  • 78
    Schmitz S, Cherny SS, Fulker DW 1998 Increase in power through multivariate analyses. Behav Genet 28: 357363.
  • 79
    Boomsma DI, Dolan CV 1998 A comparison of power to detect a QTL in sib-pair data using multivariate phenotypes, mean phenotypes and factor scores. Behav Genet 28: 329340.
  • 80
    Hauser ER, Boehnke M, Guo SW, Risch N 1996 Affected-sib-pair interval mapping and exclusion for complex genetic traits: Sampling considerations. Genet Epidemiol 13: 117137.
  • 81
    Ott J 1991 Analysis of Human Genetic Linkage. Johns Hopkins University Press. Baltimore, MD, U.S.A
  • 82
    Zhang Y, Proenca R, Maffei M, Barone M, Leopold L, Friedman JM 1994 Positional cloning of the mouse obese gene and its human homologue. Nature 372: 425432.
  • 83
    Steel KP, Brown SDM 1994 Genes and deafness. Trends Genet 10: 428435.
  • 84
    Dietrich WF, Copeland NG, Gilbert DJ, Miller JC, Jenkins NA, Lander ES 1995 Mapping the mouse genome: Current status and future prospects. Proc Natl Acad Sci U S A 92: 1084910853.
  • 85
    Lyon MF, Rastan S, Brown SDM 1996 Genetic Variants and Strains of the Laboratory Mouse, 3rd Ed. Oxford University Press, New York, NY, U.S.A
  • 86
    Hogan B, Beddington R, Constantini F, Lacy E 1994 Manipulating the Mouse Embryo: A Laboratory Manual. Cold Spring Harbor Laboratory Press, Plainview, NY, U.S.A
  • 87
    Klein RF, Mitchell SR, Phillips TJ, Belknap JK, Orwoll ES 1998 Quantitative trait loci affecting peak bone mineral density in mice. J Bone Miner Res 13: 16481656.